Dual-energy X-ray inspection of chicken fillets containing rib bone fragments

Summary

This submission contains X-ray projections of chicken fillets containing rib bone fragments.
Projections are acquired under different system settings: X-ray tube voltage and exposure time.
These datasets can be used for training and testing deep learning methods for foreign object detection.

The data are made available as part of the paper "X-ray Image Generation As A Method Of Performance Prediction For Real-Time Inspection: A Case Study".

Data Acquisition

The data were acquired at the FleX-ray laboratory of the Centrum Wiskunde & Informatica (CWI) in Amsterdam, the Netherlands (details can be found in [Coban 2020]).
The acquisition was performed with a planarar detector with an area of 143mm x 114mm, the projection size was 956px x 760px with a resolution of 150 μm.
The Source-to-Object distance was 990cm and the Source-to-Detector was 1059cm leading to a small magnification of 1.07 and close to the maximum field of view.
Every image was performed with a voltage of 40kV (40W power) and 90kV (45W power) to enable dual-energy analysis.

There are two datasets in the submission:
The first dataset features a variety of chicken fillets (14 different pieces) and rib bones (44 different fragments) with a size ranging from 1.5mm to 11mm.
The second dataset was made with two chicken fillets and two small bone fragments (2mm and 3mm) that were placed in different regions of the same fillet.

The first dataset was acquired with the exposure time of 1s.
In the second dataset, images of the same object were made with 1s, 100ms, 50ms, and 20ms of exposure time.

Data Description

The submission is split into "Dataset #1" and "Dataset #2"
Each dataset contains separate folders corresponding to the same acquisition settings. They are indicated in the name of the folder (e.g. 90kV_45W_100ms_1avg). Within one dataset, all folders feature images of the same objects under different settings.

Every folder is further split into training, validation, and test subsets (Dataset #2 includes only test data).

Each subset contains X-ray images ("NoBone" and "RibBone"), ground-truth segmentation ("NoBone_segm" and "RibBone_segm"), and additional information about samples ("NoBone.csv" and "RibBone.csv").
Every X-ray image is after pre-processing: flatfield and darkfield correction, and logarithm according to Beer's law.
The ground-truth contains automatically segmented chicken fillets and manually segmented bone fragments.
The file with additional information provides the ID of fillets and bone fragments, the size of the bone in mm, and the dual-energy contrast value.

Additional Links

These datasets are produced by the Computational Imaging group at Centrum Wiskunde & Informatica (CI-CWI).
The code to use these data can be found on Github: https://github.com/vandriiashen/pod2settings

Contact Details

For more information or guidance in using these datasets, please get in touch with
- vladyslav.andriiashen [at] cwi.nl

References

[Coban 2020] S. B. Coban, F. Lucka, W. J. Palenstijn, D. Van Loo, and K. J. Batenburg, “Explorative imaging and its implementation at the FleX-ray Laboratory,” J. Imaging, vol. 6, no. 18, 2020, doi: 10.3390/jimaging6040018.